Pika

View saved version →

---
name: onboarding
description: Run right after Pika is installed or when the user chooses Set up for Pika — checks the Pika MCP server is connected and signed in, opens the app gallery and suggests a first creation.
---

# Set up Pika

Pika works only through the tools of the Pika MCP server this plugin installs (`pika`, or `pika-staging` for the staging plugin): one tool per creation app (`create_video`, `headshot`, `product_ad`, …) plus `browse_apps`, `ask_for_media`, `get_job`, `get_account` and the rest. Never stand in for them with built-in image, video or audio generation.

1. If there are no Pika tools in your tool list, Pika isn't connected or signed in. Tell the user how to sign in, then stop:
   - Codex CLI: run `codex mcp login pika` (`codex mcp login pika-staging` for staging; `codex mcp list` shows the server's name and whether it's logged in).
   - Codex or ChatGPT app: sign in to Pika from the plugin's settings, then start a new chat.
   - Claude Code: run `/mcp` and sign in to Pika.
2. Call `get_account` with `{}`.
   - If it fails with a sign-in error (`SIGN_IN_REQUIRED` or "Sign in to Pika to continue"), the user isn't signed in yet. Most apps show their own sign-in prompt for it (a sign-in sheet in ChatGPT and Codex, a Connect card in Claude); ask the user to finish signing in there, then call `get_account` again. Where no prompt appears, use the steps in step 1.
   - If it returns the account, tell the user their credit balance in one line. If its summary says "No approval mode chosen yet", call `show_settings` with `{}` and follow its `next`. When it opens the settings panel, end your turn: the panel saves the pick itself and posts it in the chat (or the user tells you when they're done), so don't call `set_preferences` for it, and another Pika view opened in the same turn would replace the panel. When it lists the modes as text instead, show them one line each and save the user's reply with `set_preferences` (autopilot also needs `auto_approve_credits`).
3. Once an approval mode is saved, call `browse_apps` with `{}` to open the gallery.
4. Offer three starters, one line each:
   - Turn a product photo into a video ad (`product_ad`).
   - Make a studio headshot from a selfie (`headshot`).
   - Animate a photo into a short clip (`create_video`).
5. If the user installed Pika in the middle of a task, skip the starters and continue that task with the `pika-create` playbook.

View saved version →

---
name: pika-ad
description: Use when the user wants a video ad, commercial or promo for a product or brand from a product photo, logo or screenshot.
---

# Product or brand ad

1. Use the `product_ad` tool: it makes the whole ad, up to 30 s, in one take.
2. Gather in one message:
   - the product photos as `product_image_urls` (1–8) via `ask_for_media` with `kind: "image"` — for a brand with no physical product, a logo or app screenshot works;
   - the user's `brief`, and the screenplay you write from it as `concept`: a `logline`, `beats` (each an `action`, with optional `timecode` and `voiceover`) and an optional `cta`, which needs `audio.voiceover` or `audio.talking` on;
   - `length_s` (`6`, `15` or `30`, default 15) and the tool's other fields, filled from what the user said;
   - optionally an on-screen character with a `character.name`: the user's photo as `character.image_url`, or a stock one from `list_presets` for `product_ad` with `collection: "characters"`, its id as `character.preset` (then leave out `image_url`);
   - optionally a look: a `looks` id from `list_presets` with `collection: "looks"`, or the user's own words, as `look`.
3. Call `product_ad` with the input and a fresh `request_id` and no `max_credits`, then follow the quote's `next` (it applies the user's approval mode). On a yes, call it again with the same input and `request_id` and the agreed `max_credits`.
4. Offer follow-ups: `pika_soundtrack` to score it, `upscale`, or `extend_video` for a longer cut. For a multi-scene spot, use the `pika-short-film` playbook with the ad clips.

View saved version →

---
name: pika-add-subject
description: Use when the user wants to add a person, creature or object into an existing video (Pikadditions), not replace one.
---

# Add a subject to a video

1. Use the `add_subject_to_video` tool. To replace a character, outfit, location or product instead, use `swap_anything` (see `pika-swap`).
2. Get the `video` (`ask_for_media` with `kind: "video"`, or an earlier output url) and, optionally, an `image` of what to add, in one message.
3. Write the `prompt` (up to 4000 chars): what to add, where it is, and what it does in the scene.
4. Set `duration_s` (`5` or `10`, default 5) and `resolution` (`720p` or `1080p`, default 1080p).
5. Call `add_subject_to_video` with a fresh `request_id` and no `max_credits`, then follow the quote's `next` (it applies the user's approval mode). On a yes, call it again with the same input and `request_id` and the agreed `max_credits`.

View saved version →

---
name: pika-audio
description: Use when the user wants music, a song with lyrics, a sound effect, or a soundtrack scored to a video clip.
---

# Music, sound effects and soundtracks

1. To score a video, use `pika_soundtrack`: the clip as `video` (`ask_for_media` with `kind: "video"`, or an earlier output url) and an optional `instruction` (blank scores the picture on its own). Skip to step 4.
2. Otherwise use `create_audio` and pick `model` (default `pika/pika-audio/pika-music`); its values are the choices, listed here:
   - Music: `pika/pika-audio/pika-music` (`prompt`, `duration` 10–360 s in 5 s steps, default 30, optional `lyrics`) · `minimax/minimax-music-3.0/text-to-audio` (`prompt` and/or `lyrics`, `is_instrumental`; an instrumental needs a prompt) · `sonilo/sonilo-v1.1-music/text-to-music` (`prompt`, `duration` 5–360 s in 5 s steps) · `google/lyria-3-pro/text-to-audio` and `google/lyria-2/sound-effects` (`prompt` only).
   - Sound effects: `pika/pika-audio/pika-sfx` (`prompt`, `duration` 1–20 s, default 10) · `elevenlabs/eleven-text-to-sound-v2/sound-effects` (`text` up to 450 chars, `duration_seconds` 1–30, default 5, `prompt_influence` 0–1) · `sonilo/sonilo-v1.1-sfx/sound-effects` (`prompt`, `duration` 1–180 s).
   - Other: `bytedance/seed-audio-1.0/text-to-audio` (`prompt` up to 3000 chars).
3. Describe genre, mood, instruments and tempo for music; the source, action and space for a sound effect.
4. Call the tool with a fresh `request_id` and no `max_credits`, then follow the quote's `next` (it applies the user's approval mode). On a yes, call it again with the same input and `request_id` and the agreed `max_credits`.
5. To lay it under a video, add it to a project as an `a2` clip, the score lane (see `pika-short-film`).

View saved version →

---
name: pika-camera-director
description: Use when the user wants to reshoot or reframe a video with a new camera angle, framing or move, or asks what Camera Director can do / its presets.
---

# Camera Director

1. Use the `camera_director` tool. A shot combines a `framing`, an `angle` and a `motion`; pass `""` or leave a field out to keep the original, and at least one must be set.
2. To show the options with thumbnails, call `list_presets` for `camera_director` with `collection` `framings`, `angles` or `motions` (one per turn), and pass the picked id as `framing`, `angle` or `motion`. The values, for a summary or when the picker can't show:

   **Framing**: `extreme-wide` Extreme wide · `wide-master` Wide master · `full-shot` Full shot · `cowboy` Cowboy shot · `medium` Medium · `tight-close` Close-up · `extreme-close` Extreme close · `macro` Macro · `insert-detail` Insert detail

   **Angle**: `eye-level` Eye level · `aerial` Aerial · `overhead` Overhead · `high-angle` High angle · `low-angle` Low angle · `worms-eye` Worm's-eye · `pov` POV · `over-shoulder` Over shoulder · `profile` Profile · `through-frame` Through frame · `reflection` Reflection

   **Motion**
   - Held: `locked` Locked off · `tripod-breathe` Subtle breathe · `handheld-doc` Handheld · `handheld-urgent` Run-and-gun · `steadicam` Steadicam float · `gimbal-glide` Gimbal glide
   - Push and zoom: `push-in` Slow push-in · `pull-back` Slow pull-back · `crash-zoom` Crash zoom · `creep-zoom` Creeping zoom · `snap-zoom-out` Snap zoom out · `dolly-zoom` Dolly zoom
   - Track and orbit: `track-left` Track left · `track-right` Track right · `orbit-half` Half orbit · `orbit-full` Full 360 orbit · `follow-behind` Follow behind
   - Rise and fly: `crane-up` Crane up reveal · `crane-down` Crane down · `jib-sweep` Jib sweep · `pedestal-rise` Pedestal rise · `drone-flyover` Drone flyover · `drone-pullaway` Drone pull-away
   - Pan, tilt, focus: `whip-pan` Whip pan · `slow-pan` Slow pan · `tilt-up` Tilt up reveal · `tilt-down` Tilt down · `rack-focus` Rack focus · `focus-breathe` Soft focus drift · `roll` Slow roll
   - Signature: `snorricam` Snorricam · `bullet-time` Bullet time · `oner` Hollywood oner

3. Not every combination fits. The angle must suit the framing: tight frames drop aerial, overhead and worm's-eye; `extreme-close` takes only eye level or profile; `macro` only eye level or overhead; wides can't go over the shoulder. The motion must suit both: close and macro frames drop big traveling moves (drones, cranes, full orbit, follow, oner); `pov` keeps to head-height moves (held, push-in, pan, tilt); `over-shoulder`, `through-frame` and `reflection` keep to held, push, zoom, pan and focus moves (`over-shoulder` also tracks and tilts). An invalid combination is refused before any price with the allowed values; pick one of them and retry.
4. Get the video: `ask_for_media` with `kind: "video"`, `prepare_upload`, or an earlier output url. It must be one shot of 4–30 seconds (`gemini`: at most 10); cut a longer clip with `video_trim` first and pass its output url.
5. Use `model: "seedance"` unless the user asks for `gemini`, and set `resolution` to match the source. Put any scene direction in `notes`.
6. Call `camera_director` with a fresh `request_id` and no `max_credits`, then follow the quote's `next` (it applies the user's approval mode). On a yes, call it again with the same input and `request_id` and the agreed `max_credits`.
7. Offer follow-ups: another camera move, `extend_video` to continue the shot, or `pika_soundtrack` for music.

View saved version →

---
name: pika-character
description: Use when the user wants a reusable character — a character reference sheet, persona or avatar saved to their Library — from a photo, a description or picked traits.
---

# Character reference sheet

1. Use the `character_creator` tool: it makes a five-view reference sheet and files it as a Library character. Every call needs a `name`.
2. Pick `mode` from what the user has:
   - `image`: one front-facing photo in `image_urls` (`ask_for_media` with `kind: "image"`), optional `backstory`;
   - `describe`: a `description` plus a `style`;
   - `traits`: `characterType`, `gender`, `origin`, `skin`, `eyes`, `age`, `fashion`; each defaults (human, female, european, tan, brown, adult, casual).
3. Style and trait values are preset ids. To show them, call `list_presets` for `character_creator` with the field's collection — it defaults to `voices`, so always pass one: `style` → `styles`, `characterType` → `types`, `gender` → `genders`, `origin` → `origins`, `skin` → `skin_tones`, `eyes` → `eye_colors`, `age` → `ages`, `fashion` → `fashion_styles`. One picker per turn; ask about the rest in words. `characterType` and `fashion` also take the user's own words.
4. Optional `voice`, stored on the character: `list_presets` with `collection: "voices"` shows the voices; pass the picked id.
5. `model` defaults to `nano-banana-edit` for `image` and `nano-banana-2.1` otherwise; others: `gpt-image-2`, `nano-banana`, `nano-banana-pro`, `ideogram-v4`, `seedream-5-lite`, `seedream-5-pro`.
6. Call `character_creator` with a fresh `request_id` and no `max_credits`, then follow the quote's `next` (it applies the user's approval mode). On a yes, call it again with the same input and `request_id` and the agreed `max_credits`.
7. Offer follow-ups: use the sheet's output url as a cast reference in `short_film` or `create_video`, or as `character_urls` in `swap_anything`.

View saved version →

---
name: pika-create
description: Use when the user asks Pika to make or edit an image, video, music, sound effect or voiceover and no more specific Pika skill fits.
---

# Make or edit media with Pika

1. **Pick the app's tool.** Each Pika app is its own tool; choose by the task, not the output type. A purpose-built app (`headshot`, `product_ad`, `relight_media`, …) beats `create_video` or `create_image`, which fit when none does or the user names a model (pass it as `model`). If the user wants to look around, call `browse_apps` and read their pick with `get_panel`.
2. **Gather inputs in one message.** Map what the user already said into the tool's fields. For anything missing:
   - their photo, clip or audio → `ask_for_media` (pass `kind` when the field takes one kind, `app` with the app's id, and a short `purpose` such as "the photo becomes the first frame");
   - an earlier Pika result → `show_creations`;
   - a style, template, effect, voice or other choice → `list_presets` when the field's description names a collection or the user wants to see the choices; it lists published presets and built-in choices. An enum field's values are also its choices. A look the user describes works too.
   Open one panel per turn, media first; show presets once the media is in.
   When they say they uploaded or picked, call `get_panel` with that panel's `panel_id`.
3. **Price and run it.** Call it with the input and a fresh `request_id` and no `max_credits`, then follow the quote's `next` (it applies the user's approval mode). On a yes, call it again with the same input and `request_id` and the agreed `max_credits`.
4. **Follow `next`.** Each result names the next call, including when to wait on the job with `get_job`. Then offer the follow-ups the result suggests.

Jobs with no tool of their own:
- color grade a photo or clip → `relight_media` with the look in `description` (or `edit_image` for a photo);
- product shot from reference photos → `edit_image` with the photos in `reference_urls`;
- sound effect or music → `create_audio`;
- animate between a start and an end frame → `create_video` with model `pika`;
- cut a clip down before `camera_director` (one shot of 4–30 s) or `extend_video` (at most 15 s) → `video_trim`, then pass its output url;
- a talking head → `talking_head_studio` with a face image (`headshot` or `create_image` output, or the user's photo) and audio (the user's recording or `create_speech` output);
- new aspect ratio for a video → `create_video` with model `seedance-25`, the clip in `video_urls`, the target `aspect_ratio`, and a prompt to recompose the scene and extend its surroundings while keeping the subjects, action, timing and on-screen text.

To stop a running job the user no longer wants, call `cancel_job`. To find a job from earlier, call `list_jobs`; to show one, `show_creations` with its `job_id`.

View saved version →

---
name: pika-extend-video
description: Use when the user wants a clip to continue past its last frame — make it longer or show what happens next.
---

# Extend a video

1. Use the `extend_video` tool: it adds 5–8 s of new footage after the clip's last frame.
2. Get the clip as `start_video_url` (`ask_for_media` with `kind: "video"`, or an earlier output url). It must be an MP4 of at most 15 s and 50 MB; cut a longer one with `video_trim` first (see `pika-trim-video`) and pass its output url.
3. Set `duration` (5–8, default 5) and an optional `prompt` for what happens next (empty continues the scene). `generate_audio` defaults to true.
4. Call `extend_video` with a fresh `request_id` and no `max_credits`, then follow the quote's `next` (it applies the user's approval mode). On a yes, call it again with the same input and `request_id` and the agreed `max_credits`.
5. For a longer result, extend the output again, or join clips in a project (see `pika-short-film`).

View saved version →

---
name: pika-headshot
description: Use when the user wants a professional headshot, profile photo, LinkedIn photo or studio portrait from a casual photo of themselves.
---

# Studio headshot

1. Use the `headshot` tool. It takes one clear, front-facing photo of one person and either a `template` or a `prompt`, not both.
2. Ask for the photo with `ask_for_media` and `kind: "image"`, and in the same message ask what look they want.
3. Once the photo is in, use the look they describe as `prompt`, or call `list_presets` for `headshot` to show templates.
4. Call `headshot` with the input and a fresh `request_id` and no `max_credits`, then follow the quote's `next` (it applies the user's approval mode). On a yes, call it again with the same input and `request_id` and the agreed `max_credits`.
5. Offer follow-ups: another template, `edit_image` for touch-ups, `upscale` for print, or `create_video` to animate it.

View saved version →

---
name: pika-motion-transfer
description: Use when the user wants a character or person in an image to copy the movement, dance or performance from a reference video.
---

# Motion transfer

1. Use the `motion_transfer` tool: `image_url` is who performs; the motion to copy is exactly one of `video_url` or `sample`.
2. Get the character image (full or half body, framed like the motion clip; no face close-up for full-body motion) and the motion clip (one continuous take, performer visible throughout, no cuts) in one message via `ask_for_media`, `prepare_upload`, or earlier output urls.
3. For a stock motion instead, call `list_presets` for `motion_transfer` (collection `samples`) and pass the picked id as `sample`.
4. Pick `model`:
   - `kling/kling-3.0/motion-control` (default): `character_orientation` `video` follows the clip's motion up to 30 s, `image` keeps the image's framing up to 10 s; `audio` `original` or `off`; `resolution` `720p` or `1080p` (default). Fast or complex motion may distort or come out shorter.
   - `pika/pika-3.0/reference-to-video`: `resolution` `720p` (default) or `960p` take clips up to 20 s, `1440p` up to 10 s.
   An optional `prompt` adds direction.
5. Call `motion_transfer` with a fresh `request_id` and no `max_credits`, then follow the quote's `next` (it applies the user's approval mode). On a yes, call it again with the same input and `request_id` and the agreed `max_credits`.

View saved version →

---
name: pika-photo-effect
description: Use when the user wants a Pikaffects effect on a photo — cake-ify, crumble, explode, levitate and the like — or asks which effects exist.
---

# Pikaffects photo effect

1. Use the `add_photo_effect` tool: it animates a photo into a short video with a named effect.
2. Ask for the photo with `ask_for_media` and `kind: "image"` (or use an earlier output url).
3. Once the photo is in, call `list_presets` for `add_photo_effect` (collection `effects`) to show the effects, and pass the picked id as `effect`. Only ids from that list work.
4. Pass the photo in `image_urls`. Some effects take a second photo; if the tool says the effect needs another, ask for it and add it.
5. Call `add_photo_effect` with a fresh `request_id` and no `max_credits`, then follow the quote's `next` (it applies the user's approval mode). On a yes, call it again with the same input and `request_id` and the agreed `max_credits`.
6. Offer follow-ups: another effect, `pika_soundtrack` for sound, or `upscale`.

View saved version →

---
name: pika-podcast
description: Use when the user wants a video podcast, talk show, interview, debate or solo episode with on-camera hosts, or asks which podcast sets or voices exist.
---

# Video podcast

1. Use the `podcast_creator` tool. It renders one segment per call; an episode of `length` seconds (`auto`, `10`, `15`, `30`, `45`, `60`, `120`, `180`) has `ceil(length / 30)` segments, and `auto` is one segment the model sizes itself.
2. Gather in one message: the topic or the user's own script (`own_script: true` when verbatim), `format` (`conversation`, `interview`, `debate`, `solo`), host and guest names, personas and voices, and optional host and guest photos (`ask_for_media` with `kind: "image"`; without photos it is text-to-video).
3. Set: call `list_presets` for `podcast_creator` with `collection: "sets"` and pass the picked id as `set` (default `modern`).
4. Voices: `list_presets` with `collection: "host-voices"` or `"guest-voices"`, passing the pick as `host.voice` or `guest.voice`. A lowercase name also works (`adam`, the host default; `rachel`, the guest default; `sarah`, `george`, `brian`, …). A solo episode ignores the guest.
5. Write the dialogue per segment as `Host: …` / `Guest: …` lines, about 2.4 words per second of segment (a segment is the episode length divided evenly, at most 30 s), or the speech comes out rushed. Optional: `layout` (`16:9`, `9:16`, `1:1`), `quality` (`720p`, `1080p`), `scene_notes`, up to 8 `reference_urls`.
6. For more than one segment, `create_project` first and pass its `project_id` on every call. A new project is 1080x1920 and calls take its aspect unless you pass `layout`, so for `16:9` or `1:1` set its `width` and `height` with `update_timeline` first. Use one `continuity_id` for the episode and `segment_index` 0, 1, … with the same host, guest, set and length.
7. Quote each segment: call `podcast_creator` with its own fresh `request_id` and no `max_credits`. Sum the quotes into one episode total and follow the quotes' `next` with that total as the plan. On a yes, call each again with the same input and `request_id` and its quoted `max_credits`, keeping a running total within the agreed one.
8. When the segments are finished, `get_project` for the `revision` and `add_clips` them in segment order (`duration` = the segment's seconds) with a one-line `note` saying what you placed, then `export_project` with a fresh `request_id`.

View saved version →

---
name: pika-relight
description: Use when the user wants to change the lighting of a photo or clip — golden hour, studio softbox, neon, moody — or color grade it.
---

# Relight or color grade

1. Use the `relight_media` tool. Set `kind` to `photo` for an image or `clip` for a 2–30 s MP4 or MOV.
2. Get the media: `ask_for_media`, `prepare_upload`, or an earlier output url, as `media_url`.
3. Choose the light, one of:
   - a rig preset: call `list_presets` for `relight_media` (collection `rigs`) and pass the picked id as `preset`; with no preset and no lights it uses `golden-hour`;
   - `lights`: 1–3 placed lights, each with `azimuth` (-180–180; 0 front, 90 camera-right, 180 behind), `elevation` (-90–90), `color` (`#rrggbb`), `brightness` and `diffuse` (0–100); not together with `preset`;
   - a `description` of the light or grade (palette, contrast, film look), alone or with either of the above.
4. Call `relight_media` with a fresh `request_id` and no `max_credits`, then follow the quote's `next` (it applies the user's approval mode). On a yes, call it again with the same input and `request_id` and the agreed `max_credits`.
5. Offer follow-ups: another rig, or `upscale`.

View saved version →

---
name: pika-remix
description: Use when the user wants to continue from an earlier Pika result — redo, vary, edit, extend, upscale, animate or score something Pika already made.
---

# Continue from an earlier result

1. **Find the result.**
   - Made in this conversation: use its `job_id` and `outputs[].url`.
   - @-mentioned as `pika://creation/<id>`: read that resource for its outputs.
   - Made earlier or elsewhere: call `ask_for_media` with `source: "creations"`, or `show_creations` and read the pick with `get_panel`, or `list_jobs` when the user names it clearly enough to find without a gallery.
2. **Pick the tool.** Redo or vary → the same app's tool with a changed `prompt`; edit an image → `edit_image`; animate an image → `create_video`; longer video → `extend_video`; sharper → `upscale`; new lighting → `relight_media`; new outfit or object → `swap_anything`; music → `pika_soundtrack`.
3. **Pass the result as media.** Put the earlier output's URL in the field that takes that media. Keep the settings the user didn't ask to change.
4. Call it with the input and a fresh `request_id` and no `max_credits`, then follow the quote's `next` (it applies the user's approval mode). On a yes, call it again with the same input and `request_id` and the agreed `max_credits`. If the original was part of a project, pass the same `project_id`.

View saved version →

---
name: pika-resize
description: Use when the user wants a photo, poster or ad image adapted to another aspect ratio or format, with its text reflowed (Smart Resize).
---

# Smart Resize an image

1. Use the `resize_images` tool: it reframes one image to a new aspect ratio and reflows its text, keeping the copy and branding. For a video, use `pika-create` (it covers new aspect ratios for video).
2. Get the image in `image_urls` (`ask_for_media` with `kind: "image"`, or an earlier output url) and the target `aspect_ratio`: `1:1` (default), `9:16`, `16:9`, `3:4`, `4:3`, `21:9`, `2:3`, `3:2`, `4:5`, `5:4`. For several sizes, run once per ratio.
3. Optional: `resolution` `1K`, `2K` (default), `4K`; `quality` `low`, `medium`, `high` (default), `xhigh`, `max`; `output_format` `png` (default) or `jpeg`; `image_model` `openai/gpt-image-2.5-sunburst/image-to-image` (default) or `openai/gpt-image-2/image-to-image`, which caps quality at `high`; layout instructions in `prompt`.
4. Call `resize_images` with a fresh `request_id` and no `max_credits`, then follow the quote's `next` (it applies the user's approval mode). On a yes, call it again with the same input and `request_id` and the agreed `max_credits`.

View saved version →

---
name: pika-short-film
description: Use when the user wants one longer video built from several clips or scenes — a short film, story, trailer, montage or multi-scene ad.
---

# Short film from several clips

Use `short_film`; other tools fit only a one-off insert shot. Each call renders consecutive beats of the whole shot list as one continuous clip locked to the cast's stills. Make the stills with `create_image` (or `edit_image` from the user's photos) and pass their output urls as `cast[].image_url`. `style` must be a `styles` id from `list_presets`; put the user's own look in `style_reference_url` or the beats.

Timing: `length_sec` is 8, 15, 30, 60, 90 or 120; `beats` holds at most 8. Each beat runs clamp(round(length_sec / beats), 4, 15) s. A call renders `beat_count` beats from `first_beat` (0-based), at most 30 s of beats for `model` `seedance25` (default) and 15 s for `seedance` and `minimax-h3`.

1. **Plan and price it in one message.** Agree the shot list (camera, action, voiceover per beat), cast, `style` and length, then split the beats into calls that cover every beat once, in order, each within the clip limit. Quote every call (stills, each clip, the score; a clip quoted before its stills exist may omit `cast`, except on autopilot, where a call within the limit runs instead of quoting) and show one total. Follow the quotes' `next`; after a yes, run each call with its quoted `max_credits` while your running total stays within the agreed total.
2. **Create the project.** `create_project` with a name from the premise. Keep its `project_id` for the rest of the task; runs with it take the project's aspect unless you pass one. It starts at 1080x1920, so for `16:9` or `1:1` set its `width` and `height` with `update_timeline` first.
3. **Make each clip.** Every call passes the same `beats`, `cast`, `style`, `length_sec` and `project_id`, with its own `first_beat`, `beat_count` and `request_id`. To check a still or clip (the same character across shots), `get_job` with `preview: true`.
4. **Place the clips.** When the clips are finished, `get_project` for the current `revision`, then one `add_clips` with the clips in beat order (`duration` = `beat_count` × beat seconds); add the `create_audio` score as an `a2` clip with `at: 0, duration: "to_end"`. Refer to clips by the numbers `get_project` gives; the user may be editing in the Pika editor too, so always pass the latest `revision`, and pass a one-line `note` saying what you changed and why.
5. **Export when the user is happy.** `export_project` with a fresh `request_id` (it exports the current timeline). The export panel shows progress; `get_project` shows its status and the `download_url` to share.
6. Offer `pika_soundtrack` to score the finished cut. `show_projects` shows the user their projects.

View saved version →

---
name: pika-speech
description: Use when the user wants text spoken aloud — a voiceover, narration or text-to-speech clip — or asks which Pika voices exist.
---

# Speech and voiceover

1. Use the `create_speech` tool. Pick `model` (default ElevenLabs Multilingual v2); each takes different fields:

   | `model` | Script field (max chars) | Voice field |
   | --- | --- | --- |
   | `pika/pika-audio/pika-speech` | `script` (15000) | `voice_preset`, default `calm_documentary_narrator` |
   | `elevenlabs/eleven-multilingual-v2/text-to-speech` | `text` (10000) | `voice_id`, default Rachel; `stability` 0–1, default 0.5 |
   | `elevenlabs/eleven-turbo-v2-5/text-to-speech` | `text` (40000) | same as Multilingual |
   | `minimax/minimax-speech-02-hd/text-to-speech` | `text` (50000) | `voice_id`: any MiniMax voice id, free text, default `Wise_Woman` |
   | `minimax/minimax-speech-02-turbo/text-to-speech` | `text` (50000) | same as HD |
   | `bytedance/seed-audio-1.0/text-to-audio` | `prompt` (3000) | `voice`, optional; omit and the model picks |

2. To show the voices, call `list_presets` for `create_speech` with the model's collection and pass the pick in its field: `pika-voices` → `voice_preset` (76 stock voices, e.g. `deep_trailer_bass`, `polished_news_anchor`, `cozy_grandpa_storyteller`), `elevenlabs-voices` → `voice_id`, `seed-voices` → `voice`. MiniMax has no list; any MiniMax voice id works.

   ElevenLabs ids: `21m00Tcm4TlvDq8ikWAM` Rachel · `pNInz6obpgDQGcFmaJgB` Adam · `EXAVITQu4vr4xnSDxMaL` Sarah · `ErXwobaYiN019PkySvjV` Antoni · `AZnzlk1XvdvUeBnXmlld` Domi · `IKne3meq5aSn9XLyUdCD` Charlie · `JBFqnCBsd6RMkjVDRZzb` George · `TX3LPaxmHKxFdv7VOQHJ` Liam · `pFZP5JQG7iQjIQuC4Bku` Lily · `nPczCjzI2devNBz1zQrb` Brian · `cgSgspJ2msm6clMCkdW9` Jessica

3. Call `create_speech` with a fresh `request_id` and no `max_credits`, then follow the quote's `next` (it applies the user's approval mode). On a yes, call it again with the same input and `request_id` and the agreed `max_credits`.
4. Offer follow-ups: `talking_head_studio` to lip-sync a face to it (see `pika-talking-head`), or `create_audio` for music under it.

View saved version →

---
name: pika-swap
description: Use when the user wants to replace the person, outfit, location or product in an existing video while keeping its motion and camera.
---

# Swap anything in a video

1. Use the `swap_anything` tool: it reshoots the clip with Seedance 2.5, keeping the motion, camera, performance and audio. To add something new instead, use `add_subject_to_video` (see `pika-add-subject`).
2. Get the `video_url` and the reference images in one message (`ask_for_media`, `prepare_upload`, or earlier output urls). Up to 3 images each in `character_urls`, `outfit_urls`, `location_urls`, `product_urls`; at least one in total. A `character_creator` sheet works as a character.
3. Set `resolution` to match the source height (`480p`, `720p`, `1080p`, default 1080p). `restyle: true` rebuilds the whole look from the references instead of swapping per category. `generate_audio: false` mutes the output.
4. Put scene direction in `notes`; `@image N` names the Nth reference, counting characters, outfits, location, then products.
5. Call `swap_anything` with a fresh `request_id` and no `max_credits`, then follow the quote's `next` (it applies the user's approval mode). On a yes, call it again with the same input and `request_id` and the agreed `max_credits`.

View saved version →

---
name: pika-talking-head
description: Use when the user wants a photo or portrait to talk or lip-sync to a recording or script — a talking head, spokesperson or avatar video.
---

# Talking head

1. Use the `talking_head_studio` tool: one front-facing portrait and one speech track (`audio_url`) become a lip-synced video.
2. The face, exactly one of: `face_url` for the user's photo (`ask_for_media` with `kind: "image"`) or a `headshot` or `create_image` output; or `face`, a stock face id from `list_presets` for `talking_head_studio` (collection `faces`).
3. The audio: the user's recording (`ask_for_media` with `kind: "audio"`) or a script spoken with `create_speech` (see `pika-speech`), whose output url you pass once it has finished.
4. Pick `mode` by speech length: `std` Kling Standard or `pro` Kling Pro take 2–300 s; `seedance` Seedance 2.5 or `wan` Wan 3.0 take 2–30 s. Default `std`. Put the setting in `scene` (e.g. "in a sunlit kitchen").
5. Call `talking_head_studio` with a fresh `request_id` and no `max_credits`, then follow the quote's `next` (it applies the user's approval mode). On a yes, call it again with the same input and `request_id` and the agreed `max_credits`.
6. Offer follow-ups: `upscale` the video, or another script on the same face.

View saved version →

---
name: pika-trim-video
description: Use when the user wants to cut a video to a time range, or a clip is too long for another Pika tool's limit.
---

# Trim a video

1. Use the `video_trim` tool: it keeps the span from `trim_start_ms` (default 0) to `trim_end_ms`, with its audio.
2. Get the `video` (`ask_for_media` with `kind: "video"`, or an earlier output url) and the span in milliseconds; `trim_end_ms` must be after the start and at most the video's length.
3. Call `video_trim` with a fresh `request_id` and no `max_credits`, then follow the quote's `next` (it applies the user's approval mode). On a yes, call it again with the same input and `request_id` and the agreed `max_credits`.
4. When the trim feeds another tool (`camera_director`, `extend_video`, …), wait for it with `get_job` and pass its output url on.

View saved version →

---
name: pika-upscale
description: Use when the user wants a photo or video sharper or at higher resolution — upscale, enhance, 4K, print size.
---

# Upscale

1. Use the `upscale` tool with `mode: "image"` and `image_url`, or `mode: "video"` and `video_url` (`ask_for_media`, `prepare_upload`, or an earlier output url).
2. Image: `upscale_factor` `x2` (default), `x3` or `x4`, keeping the output at or under 24 megapixels; `model`:
   `standard-v2` (default) general purpose · `high-fidelity-v2` already-sharp sources · `low-resolution-v2` small, soft or compressed sources · `cgi` illustration, 3D, game captures · `text-refine` screenshots, documents, logos, UI · `wonder-3` generative new detail, costs more.
   Also `output_format` (`jpeg` default, `png`), `face_enhancement` (default on), and `sharpen`, `denoise`, `fix_compression` 0–100.
3. Video: `target_height` `1080` (default) or `2160` (4K), and `source_height` when known; `model`:
   `starlight-fast-2` (default) production default · `proteus` cheapest and fastest, tunable · `starlight-hq` clean, high-value sources · `starlight-mini` degraded or compressed footage · `starlight-sharp` soft or slightly out-of-focus footage. Starlight models expect 30 fps or below. `compression`, `noise`, `halo`, `recover_detail`, `grain` (0–100) apply to `proteus` only.
4. Call `upscale` with a fresh `request_id` and no `max_credits`, then follow the quote's `next` (it applies the user's approval mode). On a yes, call it again with the same input and `request_id` and the agreed `max_credits`.

View saved version →

---
name: prompt-gpt-image
description: Use when writing or improving a prompt for a Pika run that uses GPT Image — `create_image` or `edit_image` with model gpt-image-2.
---

# GPT Image prompts

Start from `prompt-image`; this adds what GPT Image rewards.

## In Pika

- Model id `gpt-image-2` in `create_image`. Fields besides `prompt`: `aspect_ratio`, `quality`, `background` (transparent for cut-outs and logos), `output_format`, and `image_urls` for edits and references. In `edit_image` the prompt goes in `instructions`, with `reference_urls`, `resolution` and `quality`.
- Put size, quality and transparency in their fields, not the prompt.

## Structure

- Lead with the result and its use: "A product photo for a store listing", "A poster for a jazz night".
- Describe what is visible: materials, textures, light, color, medium. Ask for "a real photograph" or "photorealistic" when that's the goal.
- For people: framing ("full body, feet visible"), gaze, pose, what they hold.
- For a long request, labeled sections read best: Scene, Subject, Details, Text, Constraints.

## Photoreal look

Describe it like a photographer: framing ("medium close-up at eye level"), lens ("50 mm, shallow depth of field"), film or grain, natural texture (pores, wear, creases), and "no heavy retouching" when it should look candid.

## Text in the image

- Exact words in quotes, with position and lettering: `the headline "OPEN LATE" in bold condensed white capitals across the top`.
- Spell unusual names letter by letter if they come back wrong.
- Ask for no other text. Check the spelling in the result; for small or dense text, try a higher `quality`.

## Edits and references

- Say "change only …", then list what must stay: identity, pose, geometry, label, lighting, colors, camera angle, background.
- Number the inputs and give each a role: "Image 1 is the person, image 2 is the jacket; dress the person in the jacket."
- One change per run; feed the result into the next edit and repeat what must be preserved.

## Examples

> A photorealistic candid photo of an elderly fisherman mending a net on a small wooden boat at dawn. Weathered skin with visible texture, a faded blue wool sweater. Medium close-up at eye level, 50 mm lens, soft low sun from the right, light film grain, no heavy retouching.

> Change only the background of image 1 to a sunlit white studio sweep. Keep the sneaker's shape, colors, laces and logo exactly, and keep the camera angle and shadows under the shoe.

> A minimal flat logo for a coffee cart named "Little Bean": a single coffee bean with a leaf, clean vector shapes, strong silhouette, two colors (espresso brown and cream), balanced negative space, no other text.

View saved version →

---
name: prompt-image
description: Use when writing or improving the prompt for a Pika image run — create, edit or restyle an image — before running it, whatever the model.
---

# Image prompts

## In Pika

- The prompt goes in the tool's `prompt` field (`instructions` in `edit_image`). A template or preset field may replace it; send one or the other.
- Check the model: `create_image` and `edit_image` take a `model`. For `gpt-image-2` also apply `prompt-gpt-image`; for Nano Banana (`nano-banana`, the default, and its variants) apply `prompt-nano-banana`.
- Keep the user's intent and words. Expand a described look into concrete visual detail; don't swap in a different idea.

## Structure

1. **Result and use** — what the image is for: product shot, poster, profile photo, thumbnail, illustration.
2. **Subject** — who or what, pose, expression, clothing, materials.
3. **Composition** — framing (close-up, full body, overhead flat lay), subject placement, background.
4. **Light** — source, direction, softness, time of day.
5. **Style** — photo (lens, depth of field, film stock) or medium (watercolor, 3D render, flat vector, anime), plus a color palette.
6. **Text**, if any — exact words in quotes, where they sit, and the lettering style.

Use plain descriptive sentences. For a busy scene, short labeled lines (Scene, Subject, Details, Text) stay readable.

## Edits

- Say what changes and list what stays: face, pose, product shape, label, background, lighting, camera angle.
- One change per run. Feed the result into the next edit for the next change.
- With several input images, give each a role: "the person from the first image, the jacket from the second".

## Leave out

- Empty praise ("beautiful, stunning, masterpiece, 8K").
- Negative lists. Describe the wanted result ("clean white background") rather than what to avoid.
- Style stacks that contradict each other ("photorealistic watercolor anime").

## Example

> Product photo for an online store: a matte black water bottle standing on pale travertine, centered, shot from slightly above at a 3/4 angle. Soft daylight from the left with a gentle shadow to the right. Shallow depth of field, neutral warm palette; the bottle is the only object in frame.

View saved version →

---
name: prompt-minimax-h3
description: Use when writing or improving a prompt for a Pika run that uses MiniMax H3 — `create_video` with model minimax-h3.
---

# MiniMax H3 prompts

Start from `prompt-video`; this adds what H3 rewards.

## In Pika

- Model id `minimax-h3` in `create_video`. Up to 15 s.
- A first frame to animate goes in `image_url` (with an optional `end_image_url`); images, clips or audio in `image_urls`, `video_urls`, `audio_urls` form a reference set mentioned as `@image1`, `@video1`, `@audio1`. Send frames or references, not both. Reference audio needs an image or clip alongside it.
- The `create_video` field descriptions give the durations, ratios and resolutions per model.

## Write it as a timeline

1. **Open with the look.** The first sentence sets the medium and style ("Live-action, cinematic, 35mm") and the opening composition.
2. **Describe in time order.** Visuals and actions flow as prose, beat by beat.
3. **Camera in plain sentences**, using three parts: the move (push in, pull out, pan, tilt, truck, tracking, arc, static, handheld shake, POV), its amplitude (small or large) and its speed (slow or fast). "The camera pushes in with small amplitude at slow speed toward the letter." Don't tack camera labels onto the end of sentences.
4. **People** — age, look and clothing on first appearance; keep those details identical afterwards.
5. **Text on screen** — the exact words in quotes.

## Multi-shot

Shot 1 has no timestamp. Each later shot starts with its cut time: "[Shot 2] At 00:05, the camera cuts to …". A cut should change the viewpoint, subject or place; a small reframing is a camera move, not a cut.

## From a first frame

When one image is animated, the image is the first frame. Describe what happens next from that exact composition — don't re-describe the image. With references, say what each one supplies (the character from @image1, the room from @image2).

## Avoid

- Abstract mood words; describe instruments, pace and visible light instead.
- Translating quoted dialogue or on-screen text.
- Changing a character's look between shots.

## Examples

Animate a portrait (first frame):

> Live-action, cinematic. The young woman at the rain-streaked train window turns her head slowly toward the camera and smiles; lights from passing towns slide across her face. The camera pushes in with small amplitude at slow speed until her eyes fill the frame.

Two shots, 8 s:

> Live-action, warm documentary style. A baker in a flour-dusted apron folds dough on a steel table at dawn; the camera trucks right with small amplitude at slow speed. [Shot 2] At 00:04, the camera cuts to a close-up of the same baker's hands sliding a tray of loaves into a glowing oven.

View saved version →

---
name: prompt-nano-banana
description: Use when writing or improving a prompt for a Pika run that uses Nano Banana (Google Gemini image) — `create_image` or `edit_image` with a nano-banana model.
---

# Nano Banana prompts

Start from `prompt-image`; this adds what Nano Banana rewards.

## In Pika

- Model ids `nano-banana` (the default), `nano-banana-edit` (needs a reference), `nano-banana-2.1` and `nano-banana-pro` in `create_image`, up to 16 reference images in `image_urls`. `edit_image` also offers `nano-banana-2`, with the photo plus up to 15 `reference_urls`.
- Put aspect ratio and resolution in their fields, not the prompt.

## Formulas

- **New image**: subject → action → place → composition → style. Write full sentences, not keyword lists.
- **With references**: the references → how they relate → the new scene. "Use the sketch for the shape and the fabric photo for the texture; render a 3D armchair in a sunlit minimalist living room."
- **Edit**: start with the verb and name only the change — "Remove the man on the left", "Make it night". Everything else is kept.
- **Combine or restyle**: a base image plus what to add from the others, or the style to carry over.

## Creative direction

- **Light**: "three-point softbox", "golden hour backlight", "high-contrast chiaroscuro".
- **Camera and lens**: "low angle, shallow depth of field (f/1.8)", "wide-angle", "macro"; a camera type sets a feel (disposable camera for snapshots, GoPro for action).
- **Color and film**: "1980s color film, slight grain", "muted teal grade".
- **Materials**: name them — "navy tweed blazer", not "suit"; "brushed steel", not "metal".

## Text

Exact words in quotes with a font description ("bold white sans-serif"). For several lines, list each line with its words and style. To get the text in another language, say which one.

## Consistency

Keep the same reference images and repeat the key identity details (face, outfit, product) in each follow-up edit. Change one thing per run.

## Avoid

- Keyword soup ("woman, red, studio, 8k, masterpiece").
- Saying what you don't want ("no cars"); say what you do ("an empty street").

## Examples

> A fashion model in a tailored brown dress and sleek boots, holding a structured handbag, stands turned slightly away against a deep cherry-red studio backdrop. Medium-full shot, centered. Fashion editorial, medium-format film, pronounced grain, rich saturation.

> Put the person from image 1 in the jacket from image 2, standing on a snowy mountain ridge at sunrise. Keep their face and hairstyle exactly. Wide shot, low angle, crisp cold light.

> A glossy beauty shot of a minimalist moisturizer jar on wet stone. Three lines of text above it: "GLOW" in flowing brush script, "10% OFF" in heavy condensed capitals, "Your first order" in a thin geometric sans.

View saved version →

---
name: prompt-seedance
description: Use when writing or improving a prompt for a Pika run that uses Seedance — `create_video` with model seedance-25, seedance or seedance-mini.
---

# Seedance prompts

Start from `prompt-video`; this adds what Seedance rewards.

## In Pika

- Model ids: `seedance-25` (Seedance 2.5: up to 30 s, native audio, up to 30 image / 10 video / 10 audio references), `seedance` (2.0: up to 15 s) and `seedance-mini`. `product_ad` (Seedance 2.5) and `short_film` (Seedance 2.5 by default) write their own prompts from their fields; this skill applies where a `prompt` field is open.
- Mention references as `@image1`, `@video1`, `@audio1`, in the order of the `image_urls`, `video_urls`, `audio_urls` you pass. The `create_video` field descriptions give the exact durations and reference limits for the route.

## Write it like production notes

For a simple shot, subject plus action is enough. For anything longer, use short labeled blocks, in this order, and skip the ones you don't need:

1. **Format** — length, ratio, live action or animation.
2. **References** — one job per reference: "@image1 is the product: keep its shape, color and label exactly. Ignore @image1's background."
3. **Opening state** — where everyone and everything is in the first frame.
4. **Timeline** — the action in time blocks ("0–5 s", "5–10 s"). Each block starts from the physical result of the one before. Cause before effect: contact, then the reaction.
5. **Camera** — shot size, angle and one movement per shot, tied to what happens ("pushes in as she lifts the cup").
6. **Continuity** — when something leaves frame and returns, repeat its key details (same face, same green coat).
7. **Audio** — dialogue in quotes with who says it, then ambient sound and music. Seedance 2.5 renders audio natively.
8. **Constraints** — only what the model gets wrong without being told ("one continuous take", "the logo stays readable").

## Multi-shot

Label each shot — `Shot 1`, `Shot 2` — each with its own framing, action and end point, and end each with "Hard cut." Keep the cast and product descriptions identical across shots.

## Avoid

- Vague motion words ("dynamic", "epic camera"): name the move.
- Several references for the same element; one clear reference per face, product, place or style works better than many.
- Cramming a 30 s story into one sentence: long clips need a timeline, or actions get skipped or rushed.

## Examples

Product, 10 s, `@image1` = product photo:

> @image1 is the espresso maker: keep its matte cobalt shell and copper button exactly. On a light oak counter in soft morning light, a hand enters from frame right and presses the button once. Dark espresso streams into the glass below; steam curls up. Close-up at counter height, slow push-in, one continuous take. Sound: a soft click, then a low hiss and pouring.

Two shots, 12 s:

> Shot 1 (0–6 s): wide shot of a rain-soaked night street, a cyclist in a yellow jacket rides toward camera through neon reflections; the camera tracks backward at her speed. Hard cut.
> Shot 2 (6–12 s): close-up of the same cyclist in the yellow jacket, rain on her face, as she brakes and looks up at a flickering sign. Static camera. Sound: rain, tire hiss, a distant tram bell.

View saved version →

---
name: prompt-video
description: Use when writing or improving the prompt for a Pika video run — text-to-video, image-to-video or a video edit — before running it, whatever the model.
---

# Video prompts

## In Pika

- The prompt goes in the tool's `prompt` field. Some apps build the prompt themselves from other fields; fill those instead.
- Check the model: `create_video` takes a `model`. For Seedance (`seedance-25`, `seedance`, `seedance-mini`) also apply `prompt-seedance`; for MiniMax H3 (`minimax-h3`) apply `prompt-minimax-h3`.
- Keep the user's intent and words. Expand a described look ("more cinematic", "like a perfume ad") into the concrete parts below; don't invent a different idea.
- References are mentioned by kind and position — `@image1`, `@video1`, `@audio1` — matching the order of `image_urls`, `video_urls`, `audio_urls`.

## Structure

Write it in this order, one or two sentences per part:

1. **Subject** — who or what, with the details that must stay the same (face, outfit, product, logo).
2. **Action** — one main action per shot, as a sequence of visible physical steps. Say the cause before the effect.
3. **Scene** — place, time of day, weather, what's in the background.
4. **Camera** — shot size (wide, medium, close-up), angle (eye level, low, overhead) and one movement (static, push in, pull out, pan, tilt, tracking, orbit, handheld, crane).
5. **Light and style** — light source and direction (golden hour backlight, soft window light, neon), color palette, medium (live action, 35mm film, 3D animation, anime).
6. **Pacing** — fit the action to the duration: a 5 s clip holds one beat; 10–15 s holds two or three.

Two to five sentences is usually enough for one shot.

## Text-to-video vs image-to-video

- **Text-to-video**: describe everything above.
- **Image-to-video**: the image already sets subject, scene and style. Describe only what moves and how the camera behaves, and name what must stay stable ("keep the label and the bottle's shape unchanged").
- **Edit or restyle a clip**: say what changes and what stays (timing, motion, framing).

## Leave out

- Mood words with nothing visible behind them ("epic", "stunning", "dynamic"): replace each with the light, color or motion that shows it.
- Several competing camera moves in one shot.
- Negative lists ("no blur, no distortion"). State the wanted result instead: "steady camera", "slow, controlled motion".
- Quality tags ("4K, masterpiece, trending"): resolution is a field, not a prompt word.

## Example

A photo of a ceramic mug as `@image1`, 5 s:

> Steam rises slowly from the mug in @image1 on a wooden cafe table. Morning sun from a window on the left throws long soft shadows. The camera pushes in gently at eye level until the mug fills the frame. Keep the mug's shape and glaze unchanged.